Development of a near-wall turbulence model and application to jet

impingement heat transfer

Tae Seon Park
a
, Hyung Jin Sung
b,
*
a
Space Propulsion Department, Rocket Technology Division, Korea Aerospace Research Institute,
52 Oun-dong, Yusong-ku, Taejon 305-333, South Korea
b
Department of Mechanical Engineering, Korea Advanced Institute of Science and Technology,
373-1 Kusong-dong, Yusong-ku, Taejon 305-701, South Korea
Received 30 March 1998; accepted 4 July 2000
Abstract
A near-wall turbulence model was developed and applied to heat and uid ows for an impinging jet ow. The kef
l
model was
modied for predictions in strongly strained turbulent ows. To derive a realizable eddy viscosity, the f
l2
function was newly
formulated. The near-wall eect and the anisotropic production were reected in the e-equation. The model performance was
validated by the relevant experimental data. The predicted results were compared with those by the ke model and by the kev
2
model. The application to jet impingement heat transfer revealed that the present numerical predictions of wall heat transfer show
good agreement with the experimental data in the stagnation region. The inuence of the Reynolds number on the stagnation
Nusselt number was investigated. The model performance was shown to be generally satisfactory. 2001 Elsevier Science Inc. All
rights reserved.
1. Introduction
Jet impingement heat transfer is used in many engineering
and industrial applications. These include cooling of electronic
components, drying paper, textiles and tempering of glass.
This technique is very attractive and cost-eective because it
can increase heat ux signicantly near the stagnation point
(Behnia et al., 1998). The impinging jet ow, despite its rela-
tively simple geometry, exhibits extremely complex ow char-
acteristics. Among others, the ow in stagnation region is
nearly irrotational and there is a large total strain along the
streamline. Also, it turns in the radial direction with substan-
tial curvature. Away from this region, the ow forms wall jet
boundary layers along the plate. An accurate prediction of
ow and attendant heat transfer in the jet impingement poses a
signicant problem.
Comprehensive knowledge of ow structure is an essential
building block to analyze the attendant heat transfer. To
predict a jet impingement heat transfer accurately, reliable ow
computations with an appropriate turbulence model should be
preceded. A literature survey reveals that many turbulence
models have been assessed for jet impingement heat transfer.
Four turbulence models were tested by Craft et al. (1993a,b)
with relevant experimental data (Baughn and Shimizu, 1989;
Baughn et al., 1991; Cooper et al., 1993). They employed one
ke eddy viscosity model and three second moment closure
models for turbulence model assessments. Their numerical
predictions indicated that the ke turbulence model shows a
substantial overprediction in the stagnation region of an im-
pinging jet, while some second moment closure models present
better predictions near the stagnation region. Behnia et al.
(1998) applied the kev
2
model to jet impingement heat
transfer. Their numerical predictions gave much better agree-
ment with experimental data, especially the wall heat transfer
in the stagnation region.
Recently, a new kef
l
model has been developed by Park
and Sung (1997). In their model, the near-wall eect without
reference to distance and the non-equilibrium eect were in-
corporated. The non-local near-wall eect was taken into ac-
count by an elliptic relaxation approximation for the wall
damping function. The local anisotropy in strongly strained
turbulent ows was also introduced in the e-equation. It was
shown that the eddy viscosity plays an important role in the
prediction of shear ows away from the wall. These model
characteristics can improve the predictions of jet impingement
ow and heat transfer. In particular, the near-wall heat
transfer of strongly strained turbulent ows in the stagnation
region is well predicted by the mean strain dependent damping
function (Craft et al., 1996).
In the present study, a modied kef
l
model is developed
and applied for predicting an axisymmetric impinging jet ow,
where a fully developed turbulent jet is perpendicular to the
at plate with uniform heat ux. The variation of C
l
was
achieved with the aid of the CayleyHamilton theorem (Park,
1999). The near-wall eect and the anisotropic production
International Journal of Heat and Fluid Flow 22 (2001) 1018
www.elsevier.com/locate/ijh
*
Corresponding author.
E-mail address: hjsung@kaist.ac.kr (H.J. Sung).
0142-727X/01/$ - see front matter 2001 Elsevier Science Inc. All rights reserved.
PII: S 0 1 4 2 - 7 2 7 X ( 0 0 ) 0 0 0 6 9 - 2
were also reected in the e-equation. The model validation was
made by comparing the predicted results with those by the ke
model and the kev
2
model (Behnia et al., 1998). Proles of
the normalized velocity and distributions of the local wall heat
transfer coecient were analyzed for various Reynolds num-
bers (Re
D
= 23 000, 50 000 and 70 000). The predicted results
were compared with the experimental data (Baughn and Shi-
mizu, 1989; Baughn et al., 1991; Cooper et al., 1993; Yan,
1993). The predicted results were shown to be generally satis-
factory.
2. Low-Reynolds number kef
l
model
2.1. Governing equations
For an incompressible turbulent ow, the governing equa-
tions are written in Cartesian tensor notation as
oU
i
ox
i
= 0Y (1)
U
j
oU
i
ox
j
=
1
q
oP
ox
i

o
ox
j
m
oU
i
ox
j
_
u
i
u
j
_
Y (2)
where U
j
and u
j
are the jth components of the mean and
uctuating velocities, respectively, P the mean pressure, q and
m are the uid density and kinematic viscosity. In the kef
l
model of Park and Sung (1997), the unknown Reynolds stress
u
i
u
j
can be expressed in a conventional form as
u
i
u
j
= m
t
oU
i
ox
j
_

oU
j
ox
i
_

2
3
kd
ij
(3)
m
t
= C
l
f
l
k
2
e
(4)
U
j
ok
ox
j
=
o
ox
j
m
_ _

m
t
r
k
_
ok
ox
j
_
P
k
e (5)
U
j
oe
ox
j
=
o
ox
j
m
_ _

m
t
r
e
_
oe
ox
j
_
P
1
e
P
2
e
P
3
e
P
4
e
CX (6)
In the above equations, C
l
, r
k
and r
e
are the model constants.
The production of turbulent energy P
k
is dened as
P
k
= u
i
u
j
oU
i
aox
j
. The damping function f
l
is obtained by
solving the elliptic f
w
equation, which introduces the eects of
wall-proximity and non-equilibrium away from the wall. In the
e-equation, P
1
e
, P
2
e
, P
3
e
, P
4
e
and C represent the mixed produc-
tion, production by mean velocity gradient, gradient produc-
tion, turbulent production and destruction in sequence,
respectively (Park and Sung, 1997). The variations of C
l
in the
eddy viscosity m
t
are allowed by decomposing f
l
into two parts,
i.e., f
l
= f
l1
f
l2
, where f
l1
signies the eect of wall proximity
in the near-wall region while f
l2
represents the eect of non-
equilibrium away from the wall (Park and Sung, 1995).
The wall-proximity function f
l1
is obtained by solving an
elliptic relaxation equation for the wall damping function (f
w
):
m
t
= C
l
f
l1
f
l2
k
2
e
Y (7)
f
l1
= 1
_
C
W 1
exp[ (R
t
aC
W 2
)
2
[R
3a4
t
_
f
2
w
Y (8)
o
2
f
w
ox
j
ox
j
=
R
t
A
2
L
2
(f
w
1)X (9)
Here, C
W 1
, C
W 2
and A are the model constants and L is a
turbulence length scale. The turbulent Reynolds number R
t
is
dened as R
t
= k
2
ame. Note that the argument of Eq. (9) is very
similar to the conception of Durbin (1993, 1996). However, its
physical meaning is dierent. Details regarding the derivation
of f
w
equation can be found in Park and Sung (1997).
As the wall is approached, R
t
decreases rapidly and L is
reduced. The energy-containing and energy-dissipating eddies
have nearly the same order of magnitude in the vicinity of the
wall so that the inuences of low turbulence Reynolds number
and viscosity are strong. To account for these eects, a length
scale related to the damping function should be modeled. A
turbulence length scale L is adopted as (Durbin and Laurence,
1996)
L =

k
3
e
2
C
2
L
m
3
e
_ _
1a2

Y (10)
where the viscous layer close to the wall is represented by the
Kolmogorov small eddies, while the region away from the wall
is represented by the energy-containing eddies. Since A and L
in Eq. (9) represent the inuence of wall proximity, the larger
value of A and L enhances the eect of wall damping and it
yields a thick viscous sublayer. The model constants are set:
C
l
= 0X09; C
W 1
= 15; C
W 2
= 120; A = 2X4; C
L
= 70 (Durbin
and Laurence, 1996; Park, 1999).
Next, the f
l2
function is newly formulated, which describes
the variation of C
l
away from the wall. The derivation of f
l2
starts from the Reynolds stress (s
ij
= u
i
u
j
) transport equation
in a homogeneous ow,
Ds
ij
Dt
= s
ik
oU
j
ox
k
s
jk
oU
i
ox
k
U
ij
e
ij
Y (11)
where U
ij
and e
ij
denote the pressurestrain correlation and the
dissipation rate tensor, respectively. In a local equilibrium
state, Db
ij
aDt = 0 where b
ij
is the anisotropy tensor
b
ij
= s
ij
a(2k) d
ij
a3. A general algebraic form of b
ij
is ob-
tained by employing the model of Speziale et al. (1991) and the
Kolmogorov assumption e
ij
= 2ed
ij
a3,
Notation
D jet diameter
ER expansion ratio of backward-facing step
h wall heat transfer coecient, h = q
w
a(HH
j
)
H jet-to-plate distance
H
b
height of backward-facing step
k turbulent kinetic energy
k
f
thermal conductivity
Nu Nusselt number, Nu = hDak
f
Pr Prandtl number
q
w
constant surface heat ux
r radial distance
Re
D
Reynolds number based on bulk velocity,
Re
D
= U
b
Dam
U mean velocity
U
b
jet bulk velocity
X
R
reattachment length
Greeks
d
ij
Kronecker delta
H mean temperature
H
j
jet temperature
e dissipation rate of turbulent kinetic energy
j von Karman constant
T.S. Park, H.J. Sung / Int. J. Heat and Fluid Flow 22 (2001) 1018 11
b
ij
2g = C
3
_

4
3
_
S
+
ij
C
4
( 2) b
ik
S
+
jk
_
b
jk
S
+
ik

2
3
b
mn
S
mn
d
ij
_
C
5
( 2) b
ik
W
+
jk
_
b
jk
W
+
ik
_
Y (12)
where S
+
ij
= 0X5(U
iYj
U
jYi
)kae, W
+
ij
= 0X5(U
iYj
U
jYi
)kae and
g = C
1
a2 P
k
ae 1, respectively. The model constants C
1
and
C
3
C
5
are employed from the pressurestrain model of
Speziale et al. (1991).
In a matrix form, Eq. (12) can be written as
b = a
1
S a
2
bS
_
Sb
2
3
bSI
_
a
3
Wb ( bW)Y (13)
where the model constants are summarized as:
a
1
= (C
3
4a3)a2g; a
2
= (C
4
2)a2g; a
3
= (C
5
2)a2g. {.}
denotes the trace and I is the identity vector. For a two-
dimensional ow, an explicit expression of b is available in
linearly independent tensors that may be formed with S and W
(Gatski and Speziale, 1993). The general form of algebraic
solution is
b = b
1
S b
2
S
2
_

1
3
S
2
I
_
b
3
SW ( WS)X (14)
To determine the coecients b
1
b
3
, the substitution of Eq. (14)
into Eq. (13) gives
b = a
1
S 2a
2
b
1
S
2
_

1
3
S
2
I
_
a
3
b
1
WS ( SW)
2a
2
b
2
S
3
_

1
3
S
2
IS
1
3
S
3
I
_
a
3
b
2
( a
2
b
3
) WS
2
_
S
2
W
_
a
3
b
3
2WSW
_
W
2
S SW
2
_
X (15)
All higher-order tensor combinations can be reduced with the
aid of the CayleyHamilton theorem and the following rela-
tions:
S
3
= (1a3)S
3
I (1a2)S
2
ISY
WS
2
S
2
W= 0Y
WSWW
2
S SW
2
= (1a2)W
2
IS W
2
ISX
s
ij
=
2
3
kd
ij
2kb
1
S
+
ij
2k b
2
S
+
ik
S
+
kj
_ _

1
3
S
+
mn
S
+
mn
d
ij
_
b
3
W
+
ik
S
+
kj
_
S
+
ik
W
+
kj
_
_
Y (16)
b
1
=
a
1
1
2
3
a
2
2
S
+
mn
S
+
mn
2a
2
3
W
+
mn
W
+
mn
Y (17)
b
2
= 2a
2
b
1
Y (18)
b
3
= a
3
b
1
X (19)
A nal linear model for s
ij
is
s
ij
=
2
3
kd
ij
2C
+
l
k
2
e
S
ij
Y (20)
C
+
l
=
a
1
1
2
3
a
2
2
S
+
mn
S
+
mn
2a
2
3
W
+
mn
W
+
mn
X (21)
This is an isotropic eddy viscosity model with the coecient
C
+
l
, which is a function of S
+
=

2S
+
ij
S
+
ij
_
and W
+
=

2W
+
ij
W
+
ij
_
.
Recall that C
+
l
returns to 0X09 in the standard ke model. The
damping function f
l2
is obtained by f
l2
= C
+
l
aC
l
.
Although Eq. (21) provides an excellent description of the
near-equilibrium turbulent ows (Gatski and Speziale, 1993), a
singularity arises in the case of large S
+
and W
+
. In the short-
time rapid distortion theory, as S
+
, kak
o
retains a value
of order one, where k
o
is an initial turbulent kinetic energy.
Accordingly, f
l2
shows the following asymptotic behavior, i.e.,
f
l2
~ 1aS
+
. However, the behavior of C
+
l
demonstrates
C
+
l
~ (1aS
+
)
2
as S
+
and W
+
in a homogeneous
shear ow (Park, 1999).
For a weakly strained turbulent ow, the f
l2
function of
Eq. (21) is rewritten by a Pade approximation,
f
l2
~
3a
+
1
g(g
2
0X5a
+
2
2
S
+
2
)
3g
4
0X5a
+
2
2
S
+
2
g
2
3a
+
2
3
W
+
2
(g
2
0X5a
+
2
2
S
+
2
)
X (22)
Here, the model constants have the following relations:
a
+
1
= a
1
gY a
+
2
= a
2
gY a
+
3
= a
3
gX
The above relations show the limiting behaviors, f
l2
~ O(1)
and g ~ O(1) for weak and f
l2
~ 1aS
+
and g ~ S
+
for strong
strains. It is seen that Eq. (22) shows a consistent limiting
behavior for large and small strain rates.
In an eort to make a realizable linear model for s
ij
, the
conditions of the non-negativity of normal stresses and the
Schwarz's inequality between any uctuating quantities are
imposed. This is a basic physical and mathematical principle
that prevents a turbulence model from producing non-physical
results. In a limiting ow with the generalized mean strain
tensor: S
11
= cY S
22
= c(1 a)a2, S
33
= c(1 a)a2 and
S
ij
= 0 for i ,= j, where a represents types of ow deformation;
axisymmetric contraction (a = 0), plane strain (a = 1) and
axisymmetric expansion (a = 3). With this constraint, f
l2
has
the following realizable range, i.e., 0 6f
l2
62a(3C
l
S
+
) =
1a(3C
l
c

3 a
2
_
kae). A simple function can be derived by the
correction of leading term,
f
l2
=
C
l1
C
l2
n (g
2
g
2
)(C
l
g C
l3
n
2
)
C
l3
g
4
C
l4
(gg)
2
C
l3
n
2
(g
2
g
2
C
l5
g
3
)
Y (23)
where g = C
l
f
w

2S
+
ij
S
+
ij
_
, n = 2C
l
f
w

2W
+
ij
W
+
ij
_
and
g = 1 0X2g, respectively. The introduction of f
w
is that the
wall-proximity eect in the near-wall region is signicant
compared to the strain rate eect. The model constants are
determined from the experimental data of Tavoularis and
Karnik (1989): C
l1
= 7; C
l2
= 4; C
l3
= 6; C
l4
= 13; C
l5
= 10.
To look into the performance of the model, the behaviors
of 2b
12
= C
l
f
l
S
+
and 2b
11
= C
l
f
l
S
+
are plotted in Fig. 1 for
a homogeneous shear ow and an irrotational ow, respec-
tively. A global dependency of turbulent shear stress on S
+
is
clearly displayed. The results are compared with the exper-
iment of Tavoularis and Karnik (1989). Three models are se-
lected: the models of Craft et al. (1993a,b), Gatski and Speziale
(1993) and Cotton and Ismael (1993). Three models show a
good prediction for a homogeneous shear ow. However, the
model of Cotton and Ismael shows a negative value in highly
strained ows. A correct asymptotic behavior is predicted by
the present model, which is f
l2
~ 1aS
+
for S
+
. However,
the limiting behaviors of other models are less satisfactory. It is
interesting to nd that the model of Gatski and Speziale pre-
dicts well in a homogeneous shear ow while a non-physical
prediction is exhibited in Fig. 1(b). This is attributed to the
violation of realizability in an irrotational ow.
2.2. Modeling of the e-equation
By the scaling argument, P
1
e
, P
2
e
, P
4
e
and C terms in Eq. (6)
can be modeled in a way similar to the prior models (Park and
Sung, 1997)
12 T.S. Park, H.J. Sung / Int. J. Heat and Fluid Flow 22 (2001) 1018
P
1
e
P
2
e
P
4
e
C = C
+
e1
P
k
_
C
e2
f
2
e
_
aT
C
P
(1 f
w
)mm
t
o
2
U
i
ox
j
ox
j
_ _2
Y (24)
where C
+
e1
, C
e2
and C
P
are the model constants. The turbulent
timescale is dened as
T =

(kae)
2
C
2
T
T
2
K
_
Y
where the Kolmogorov timescale is T
K
=

mae
_
as a lower
bound (Durbin and Laurence, 1996). The model constants are
C
T
= 6X0 and C
P
= 0X8. The model function f
2
in Eq. (24) is
expressed as f
2
= 1 (2a9) exp(0X33R
1a2
t
), which describes
the eect of decaying turbulence.
Although most prior models have used a constant value C
+
e1
,
the variation of C
+
e1
is sensitive to the non-equilibrium eect
(P
k
ae) and to the shear layer spreading rate (Durbin, 1993;
Park and Sung, 1995). It is found that C
+
e1
gives a signicant
inuence on the reattachment length prediction in ows over a
backward-facing step (Park and Sung, 1997). This may be
attributed to the additional production of dissipation rate by
local anisotropy. Speziale and Gatski (1997) also developed C
+
e1
based on the explicit e
ij
relation with S
+
ij
and W
+
ij
. Since their
model was calibrated for a homogeneous shear ow, the ow
prediction in the near-wall region is less satisfactory. In order
to incorporate the eect of normal strain rates, a new C
+
e1
is
derived in the present study. Based on the production terms in
the e-equation with the Rotta model,
P
1
e
P
2
e
= C
+
e1
P
k
e
k
= 2m
ou
i
ox
k
ou
j
ox
k
S
ij
= C
e1
u
i
u
j
k
eS
ij
= C
e1
2b
ij
eS
ij
Y (25)
C
+
e1
is obtained by substituting the full anisotropy tensor
(Eq. (14)) and the continuity:
C
+
e1
= C
e1
C
l
f
w
e
P
k
b
+
2
S
+
ik
S
+
kj
_ _
S
+
mn
S
+
mn
d
ij
3
_
b
+
3
W
+
ik
S
+
kj
_
S
+
ik
W
+
kj
_
_
S
+
ij
Y (26)
where
b
+
2
=
2 n
2
+
1 g
+
( ) n
2
+
4 n
2
+
1 8g
3
+
_ _
35g
2
+
Y
b
+
3
=
4 n
2
+
1 g
+
( ) g
2
+
4 n
2
+
1 12g
3
+
_ _
40g
2
+
Y
g
+
= C
l

2S
+
ij
S
+
ij
_
and
n
+
= 2C
l

2W
+
ij
W
+
ij
_
X
The model constants are determined by using the algebraic
equations of b
+
2
and b
+
3
obtained from the experimental results
and numerical optimization for a fully developed channel ow
(see Fig. 2). This accounts for the additional production of
dissipation by local anisotropy.
The turbulent diusion terms of k and e, in general, are
modeled by a gradient diusion type. The values of r
k
and r
e
are usually taken to be 1X0 6r
k
Y r
e
61X4. The model constants
employed are r
k
= 1X1, r
e
= 1X3, C
e1
= 1X5 and C
e2
= 1X9.
2.3. Modeling of the temperature equation
In predicting turbulent heat transfer, the modeling of
turbulent heat ux based on the Boussinesq approximation
is generally used. The unknown eddy diusivity of heat is
y
+
U
+
10
0
10
1
10
2
0
5
10
15
20
25
DNS
present
Re =395

y
+
k
+
0 100 200 300 400
0
1
2
3
4
5
DNS
present
Re =395

y
+

+
0 50 100 150
0
0.05
0.1
0.15
0.2
0.25
DNS
present
Re =395

Fig. 2. Comparison of the predicted U

, k

and e

.
-
2
b
1
2
0
0.1
0.2
0.3
0.4
0.5
equilibrium
experiment
DNS
Cotton and Ismael
Craft et al.
Gatski and Speziale
present
S
*
-
2
b
1
1
0 10 20 30 40 50
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Cotton and Ismael
Craft et al.
Gatski and Speziale
present
realizability limit
(a)
(b)
Fig. 1. Comparison of b
12
and b
11
. (a) Homogeneous ow; (b) irro-
tational ow.
T.S. Park, H.J. Sung / Int. J. Heat and Fluid Flow 22 (2001) 1018 13
calculated by prescribing a constant turbulent Prandtl number
Pr
t
. The governing equation of mean temperature can be ex-
pressed as
U
j
oH
ox
j
=
o
ox
j
m
Pr
_ _

m
t
Pr
t
_
oH
ox
j
_
X (27)
The assumption, i.e., Pr
t
= constant satises Pope's linear
principle of scalars in turbulent ows. However, it is revealed
that there are no universal values of Pr
t
, even in simple wall
shear ows. The direct numerical simulation data showed that
Pr
t
increases towards a wall instead of being constant (Kasagi
et al., 1992). Its value is 1.1 at the wall and 0X70X9 in the outer
region.
Many attempts have been made to get a reliable Pr
t
formula
which improves the prediction of turbulent heat transfer. A
formula, incorporated with a simple conduction model which
ts the available experimental data reasonably well for a
boundary layer without pressure gradient, has a form as fol-
lows (Kays and Crawford, 1993):
Pr
t
=
1
0X5882 0X228(m
t
am) 0X0441(m
t
am)
2
[1 exp(
5X165
mt am
)[
X
(28)
This yields a value of 1.7 at the wall and an asymptotic value of
0.85 far from the wall. The present model is tested for the eect
of Pr
t
on the local Nusselt number for impinging jet ows.
2.4. Boundary conditions
A schematic diagram of the jet impingement is shown in
Fig. 3. At the jet discharge, the ow is fully developed and
isothermal. The inlet condition is obtained by the preceding
computation of fully developed pipe ow. The upper compu-
tational domain is shown in Fig. 3, which is higher 34 times of
the jet diameter. This allows the upper boundary to be a suf-
cient distance from the wall so that it does not aect the ow
near the impingement surface. Over the remainder of the upper
boundary, an entrainment condition is applied to all variables:
properties of the entering uid satisfy the zero-gradient con-
dition except the radial velocity U = 0. At the right-hand
outow boundary, the normal velocities to the boundary are
obtained from the continuity and the Neumann condition is
applied to all variables. It is found that the right boundary
location should be larger than (8D H) to avoid the ow
distortion. The boundary at the left-hand side is a symmetric
axis.
Finally, a uniform constant heat ux is applied to the wall
to compare the predicted results with the experiments (Baughn
and Shimizu, 1989; Baughn et al., 1991; Yan, 1993). A com-
parison of the condition of constant wall temperature with that
of constant heat ux reveals that a little dierent trend of local
heat transfer is exhibited. The radial and normal velocity
components are set to be zero. The asymptotic behavior
e
w
2mkan
2
as n 0 is applied for all wall boundaries, where
n denotes the wall normal direction.
2.5. Numerical procedure
The nite-dierence equations were discretized using the
hybrid linear and parabolic approximation (HLPA) scheme
with second-order accuracy. A non-staggered variable ar-
rangement was adopted with the momentum interpolation
technique to avoid the pressurevelocity decoupling. The
coupling between pressure and velocity was achieved by the
SIMPLEC algorithm. The convergence was accelerated with a
multigrid method (Park, 1999). Non-uniform, orthogonal and
axisymmetric grid was used with a high resolution near all
solid boundaries. The grid convergence was checked by three
cases: 81 51, 141 93 and 281 93. The outcome with a
141 93 mesh was found to be satisfactory, as shown in Fig. 4.
Two values of the pipe wall thickness were examined, i.e.,
0X112D and 0X0313D, which were selected from the experiments
of Cooper et al. (1993) and Baughn and Shimizu (1989), re-
spectively. The thickness 0X0313D was used for the present
computation. The computations were implemented on a
CRAY-YMP supercomputer, and a typical CPU time was
approximately 0.8 h for a 141 93 mesh. Convergence was
declared when the maximum normalized sum of absolute re-
sidual source over all the computational nodes was less than
10
4
.
3. Results and discussion
3.1. Application to fully developed channel ow
Before proceeding further, it is important to ascertain the
generality and accuracy of the present model. The model is
tested for a fully developed channel ow for which turbulence
quantities are available from the DNS data (Moser et al.,
1999). The selected Reynolds number is Re
s
= 395. The pro-
les of mean velocity, turbulent kinetic energy and its dissi-
pation rate are displayed in Fig. 2. The results of the present
model are in good agreement with the DNS data.
In order to assess the model performance of separated and
reattaching ows, a ow over a backward-facing step is se-
lected. This ow is frequently used for benchmarking the Fig. 3. Flow conguration and computational domain.
r/D
N
u
0 2 4 6
0
60
120
180
81 X 51
141 X 93
281 X 93
H/D=2
Re
D
=23000
Fig. 4. Grid convergence test.
14 T.S. Park, H.J. Sung / Int. J. Heat and Fluid Flow 22 (2001) 1018
performance of turbulence models (Park and Sung, 1997).
Table 1 lists the computational conditions and the calculated
reattachment lengths X
R
aH
b
. The present model's results are in
excellent agreement with the experiments in the range
1X125 6ER61X667. Here, ER denotes the expansion ratio.
3.2. Realizability
It has been a signicant problem of two-equation turbu-
lence models to predict an enormously large growth of tur-
bulent kinetic energy at a stagnation point. Their poor
prediction arises from the use of the explicit algebraic stress
equation based on S
+
and W
+
. In a two-dimension ow, the
turbulent production due to the irrotational strains is dened
as
P
k
= u
2
oU
ox
v
2
oV
oy
= 2m
t
oU
ox
_ _
2
_

oV
oy
_ _
2
_
X (29)
In the case of stagnation ow, the experimental result indicates
P
k
~ 0. This is due to the fact that the intensities of normal
stresses are nearly the same and oUaox = oV aoy form the
continuity. However, as can be seen in Eq. (29), the production
gives a quadratic variation in strain rates. This leads to a
spurious generation of turbulent kinetic energy.
In the ke model, this anomaly should be prevented by
deriving a realizable eddy viscosity. There are several solvable
approaches; the rst is to introduce a physical time scale or a
length scale with the realizability constraint. Durbin (1996)
imposed a bound on the time scale in the kev
2
model. The
results of jet impingement heat transfer showed that the stag-
nation anomaly can be ameliorated by this constraint. The
second is to allow the variation of C
l
, which is dened in
Eq. (4). The value of C
l
= 0X09 was chosen on the basis of
experiments in a local equilibrium ow. However, if the non-
equilibrium eect is dominant, C
l
varies depending on the ow
conditions.
The deviation from a local equilibrium is expected in the
presence of mean deformation of turbulence elds. This is
because the energy-containing eddies are strained primarily by
the mean deformation. Following Hunt's (1992) suggestion,
any mean deformations fall in the range 1 6p 61: p = 1
corresponds to a pure rotation, p = 0 a pure shear and p = 1 a
pure strain, where p = (S
ij
S
ij
W
ij
W
ij
)a(S
ij
S
ij
W
ij
W
ij
). The
ows with a strong strain (p 1) have a signicant history
eect and turbulence models based on an eddy viscosity con-
cept for Reynolds stresses lead to non-physical predictions in
even the mean ow. An example is the ke model application
to a strain-dominated impinging jet ow, where eddy viscosity
formula with its inadequate presentation of the anisotropy of
the normal stresses overpredicts the energy production terms.
As shown in Fig. 5, the deformation of ow is siginicant in
the vicinity of impinging region. To avoid this erroneous be-
havior, a dierent value of C
l
should be taken or C
l
should be
varied depending on ow deformations. In this region, the
normal stress (u
2
= 2ka3 2m
t
S
11
) of the standard ke model
represents negative values. This implies that the constant val-
ues of C
l
= 0X09 are not appropriate in the eddy viscosity
equation. In the present study, the f
l2
function of Eq. (23)
enforces the realizability for Reynolds stresses by imposing the
mean strain rate variation. As seen in Fig. 6, the inuence of
f
l2
is signicant in the impinging region. The prediction with
f
l2
= 1 is similar to the result of the ke model. This exem-
plies a closer interrelation between the value of C
l
and the
ow deformation. The present model with the realizability for
irrotational strains shows a better agreement.
The predicted distributions of turbulent kinetic energy (k)
and eddy viscosity (m
t
) near the stagnation region are com-
pared in Fig. 7 for two cases, i.e., Re
D
= 23 000 and
Re
D
= 70 000 at HaD = 2. A closer inspection of the proles of
k and m
t
reveals that f
l2
prevents the excessive high turbulence
Table 1
Comparison of X
R
with experiments
Case ER Re
H
b
Grid X
R
aH
b
(Exp.) X
R
aH
b
Driver and Seegmiller 1.125 38 000 233 201 6X21 0X2 6.31
Adams et al. 1.25 26 000 233 181 6X67 0X3 6.46
Kim et al. 1.50 46 000 233 141 7X60 0X3 7.58
Eaton and Johnston 1.667 38 000 233 101 7X95 0X3 7.97
-1 -0.6 -0.2 0.2 0.6 1
(deformation parameter)
positive
negative
u
2
/U
2
b
Fig. 5. Realizability of standard ke model.
T.S. Park, H.J. Sung / Int. J. Heat and Fluid Flow 22 (2001) 1018 15
level in the stagnation region. The ke model and the present
model with f
l2
= 1 show highly diusive results in the stag-
nation region, while the kev
2
model and the present model
predict the realizable turbulent energy and eddy diusion. This
suggests that the present f
l2
model works well in suppressing
the turbulent diusivity in the near-wall region of strongly
strained turbulent ow. As Re
D
increases, the peak of k in the
near-wall region decreases.
Next, the proles of normalized mean velocity are shown in
Fig. 8. The predicted results by three turbulence models are
compared with the experimental data of Cooper et al. (1993).
Near the stagnation region, all predictions are consistent with
the experiment. In the wall jet boundary layer where the ow is
accelerated, the ke model shows an underprediction in the
wall layer and an overprediction in the outer region. This de-
viation is attributed to the wall jet development from highly
diusive stagnation ow. However, the kef
l
and the kev
2
models resolve the behavior well, because the stresses of two
models are realizable in the stagnation region.
3.3. Heat transfer
When an eddy viscosity model is used in the prediction of
heat transfer, the most important factor in determining the
local transport is the distribution of turbulent thermal diu-
sivity (a
t
= m
t
aPr
t
). In order to evaluate the eect of Pr
t
, two
constant values and an empirical formula of Kays and Craw-
ford (1993) are employed. For HaD = 2 and Re = 23 000, the
predicted results are compared with the relevant experimental
data (Baughn and Shimizu, 1989; Baughn et al., 1991; Yan,
1993). As shown in Fig. 9, the results are in good agreement
with the measurements. Note that the formula of Kays and
Crawford (1993) gives a somewhat better agreement in the
stagnation region. However, a slight underprediction is seen in
the downstream (raDP1X5). In general, as Pr
t
decreases, the
wall heat transfer increases. A good agreement with the ex-
periment is seen at Pr
t
= 0X9.
The Nu distributions for HaD = 2 and Re
D
= 23 000 are
represented in Fig. 10. The predicted results are compared with
r/D
N
u
0 1 2 3 4 5
0
50
100
150
200
250
Baughn and Schimizu
Baughn et al.
Yan
f

k
present
k
(Park and Sung)
(Park)
H/D=2
present
( f =1)

2
Re=23000
Fig. 6. Comparison of kef
l
models with experiments.
Fig. 7. Model predictions of k and m
t
.
0 0.5 1
0
0.125
0.25
0.375
0.5
x
/
D
U/U
b
r/D=0
0 0.5 1
U/U
b
r/D=1.0
Cooper et al.
k
k v
2
Present
0 0.5 1
U/U
b
r/D=2.5
Fig. 8. Model comparisons of UaU
b
with experiment (HaD = 2 and
Re
D
= 23 000).
r/D
N
u
0 1 2 3 4 5
0
50
100
150
200
Baughn and Schimizu
Baughn et al.
Yan
Kays and Crawford
Pr
t
=0.8
Pr
t
=0.9
H/D=2
Re
D
=23000
Fig. 9. Eect of Pr
t
on Nu.
16 T.S. Park, H.J. Sung / Int. J. Heat and Fluid Flow 22 (2001) 1018
the experimental data. The results by the kev
2
model are also
displayed for comparison. This is based on the belief that the
kev
2
model can be regarded as a reliable model for predicting
the jet impingement heat transfer. As shown in Fig. 10, the
present kef
l
model follows the experimental data well. It is
evident that the stagnation heat transfer is signicantly over-
predicted by the ke model. The ke model overpredicts about
300% in the stagnation region. A closer inspection of Fig. 10
indicates that a second peak is displayed by the present kef
l
model. It is known that the second peak is associated with the
maximum turbulent kinetic energy near the wall. Within the
range of raD = 1 to 2, the Nu distribution computed with
the kev
2
model overpredicts slightly, while the present model
gives an excellent agreement with experimental data. This re-
gion is a transition zone where the stagnation ow is trans-
formed into a radial wall jet. The ring shaped wall eddies are
induced by large scale toroidal vortices hitting the plates
(Popiel and Olev, 1991). These eddies enhance local momen-
tum and heat transfer, which are responsible for the second
peak. Moreover, a second peak is clearly displayed by the
present model in Fig. 10. As mentioned earlier, since the kef
l
model includes the non-local near-wall eect and the realizable
eddy viscosity in strongly strained turbulent ows, an im-
proved performance is anticipated.
The Nu distributions for three Reynolds numbers
(Re
D
= 23 000, 50 000 and 70 000) are displayed in Fig. 11.
These results are compared with the experimental data of Yan
(1993). It is seen that some discrepancies are found between
experiment and computation. However, the discrepancy
diminishes downstream of the stagnation region. As Re
D
increases, the edge vortex of jet ow is strengthened.
4. Conclusions
In order to make a realizable stressstrain relation, the
kef
l
model has been modied with the aid of the Cayley
Hamilton theorem and the realizability constraint. The f
l2
function was introduced to derive a realizable eddy viscosity. It
prevented the spurious generation of turbulent kinetic energy
by imposing the variations of C
l
to the mean strain rates. The
eect of f
l2
on Nu was signicant in the impinging region. The
near-wall eect and the anisotropic production were reected
in the e-equation. In the rst, the present model was tested for
a fully developed channel ow and a separated and reattaching
ow over a backward-facing step. The predicted results re-
produced the wall limiting behavior successfully. The com-
puted Nu showed a good agreement with the experimental data
while the ke model overpredicted. The monotonic decrease of
Nu was attenuated near raD = 2. This behavior was due to the
insensibility of eddy viscosity in strongly strained ows. The
velocity magnitude prole was consistent with the measure-
ments in the region raDP1X5. The development of radial wall
jet for three Reynolds numbers was satisfactory. The dynamic
and thermal characteristics of impinging jet ow were well
captured by the present model.
Acknowledgements
This work was supported by a grant from the National
Research Laboratory of the Ministry of Science and Tech-
nology, Korea.
References
Baughn, J., Shimizu, S., 1989. Heat transfer measurements from a
surface with uniform heat ux and an impinging jet. ASME J. Heat
Transfer 111, 10961098.
Baughn, J., Hechanova, A., Yan, X., 1991. An experimental study
of entrainment eects on the heat transfer from a at surface to
a heated circular impinging jet. ASME J. Heat Transfer 113,
10231025.
Behnia, M., Parneix, S., Durbin, P.A., 1998. Prediction of heat transfer
in an axisymmetric turbulent jet impinging on a at plate. Int. J.
Heat and Mass Transfer 41 (12), 18451855.
Cooper, D., Jackson, D., Launder, B., Liao, G., 1993. Impinging jet
studies for turbulence model assessment-I. Flow-eld experiments.
Int. J. Heat Mass Transfer 36, 26752684.
Cotton, M.A., Ismael, J.O., 1993. Development of a two-equation
turbulence model with reference to a strain parameter. In:
Proceedings of the Fifth International Symposium on Rened
Flow Modeling and Turbulence Measurements. Paris, pp. 117124.
Craft, T.J., Graham, L., Launder, B.E., 1993a. Impinging jet studies
for turbulence model assessment-II. An examination of the
perfomance of four turbulence models. Int. J. Heat Mass Transfer
36, 26852697.
r/D
N
u
0 1 2 3 4 5
0
100
200
300
400
Baughn and Schimizu
Baughn et al.
Yan
k
v
2
k
present
H/D=2
Re
D
=23000
Fig. 10. Model comparisons of Nu with experiments.
r/D
N
u
0 1 2 3 4 5
0
50
100
150
200
250
300
350
Yan
k v
2
present
Re
D
=23000
Re
D
=50000
Re
D
=70000
H/D=2
Fig. 11. Model comparisons of Nu with experiments at three Reynolds
numbers (Re
D
= 23 000, 50 000 and 70 000).
T.S. Park, H.J. Sung / Int. J. Heat and Fluid Flow 22 (2001) 1018 17
Craft, T.J., Launder, B.E., Suga, K., 1993b. Extending the applica-
bility of eddy-viscosity models through the use of deformation
invariants and nonlinear elements. In: Proceedings of the Fifth
International Symposium on Rened Flow Modeling and Turbu-
lence Measurements. Paris, pp. 125132.
Craft, T.J., Launder, B.E., Suga, K., 1996. Development and appli-
cation of a cubic eddy-viscosity model of turbulence. Int. J. Heat
Fluid Flow 17, 108115.
Durbin, P.A., 1993. A Reynolds-stress model for near-wall turbulence.
J. Fluid Mech. 249, 465498.
Durbin, P.A., 1996. On the ke stagnation anomaly. Int. J. Heat Mass
Transfer 17, 8990.
Durbin, P.A., Laurence, D., 1996. Nonlocal eects in single point
closure. In: Third Advances in Turbulence Research Conference,
Korea Uinversity, Korea, pp. 109120.
Gatski, T.B., Speziale, C.G., 1993. On explicit algebraic stress models
for complex turbulent ows. J. Fluid Mech. 254, 5978.
Hunt, J.C.R., 1992. Developments in computational modeling of
turbulent ows. In: Pironneau, O., Rodi, W., Ryhming, I.L., Savill,
A.M., Truong, T.V. (Eds.), Numerical Simulation of Unsteady
Flows and Transition to Turbulence. pp. 176.
Kasagi, N., Tomita, Y., Kuroda, A., 1992. Direct numerical simula-
tion of passive scalar eld in a turbulent channel ow. ASME J.
Heat Transfer 114, 598606.
Kays, W.M., Crawford, M.E., 1993. Convective Heat and Mass
Transfer, third ed. McGraw-Hill, New York.
Moser, R.D., Kim, J., Mansour, N.N., 1999. Direct numerical
simulation of turbulent channel ow up to Re
s
= 590. Phys. Fluids
11 (4), 943945.
Park, T.S., 1999. Multigrid method and low-Reynolds-number ke
model for turbulent recirculating ows. Numer. Heat Transfer Part
B: Fundamentals 36, 433456.
Park, T.S., Sung, H.J., 1995. A nonlinear low-Reynolds-number ke
model for turbulent separated and reattaching ows (I) ow eld
computation. Int. J. Heat Mass Transfer 38 (14), 26572666.
Park, T.S., Sung, H.J., 1997. A new low-Reynolds-number kef
l
model for predictions involving multiple surfaces. Fluid Dyn. Res.
20, 97113.
Popiel, C.O., Olev, T., 1991. Visualization of a free and impinging
round jet. Exp. Therm. Fluid Sci. 4, 253264.
Speziale, C.G., Gatski, T.B., 1997. Analysis and modelling of
anisotropies in the dissipation rate of turbulence. J. Fluid Mech.
344, 155180.
Speziale, C.G., Sarkar, S., Gatski, T.B., 1991. Modeling the pressure
strain correlation of turbulence: an invariant dynamical systems
approach. J. Fluid Mech. 227, 245272.
Tavoularis, S., Karnik, U., 1989. Further experiments on the evolution
of turbulent stresses and scales in uniformly sheared turbulence.
J. Fluid Mech. 204, 457478.
Yan, X., 1993. A preheated-wall transient method using liquid crystals
for the measurement of heat transfer on external surfaces and in
ducts. PhD Thesis, University of California, Davis.
18 T.S. Park, H.J. Sung / Int. J. Heat and Fluid Flow 22 (2001) 1018